[IEEE 2014 International Conference on Optimization, Reliabilty, and Information Technology (ICROIT)...

6
2014 International Conference on Reliability, Optimization and Information Technology - ICROIT 2014, India, Feb 6-8 2014 Hybrid ECG Signal Compression System: A Step Towards Efficient Tele-Cardiology Rashima Mahajan, Dipali Bansal Research Scholar, Dept. ofEEE, FET, Manav Rachna Inteational University, Faridabad, India Adjunct Faculty, Dept. of ECE, SE T, Apeejay Stya University, Gurgaon, India Professor, Dept. ofEEE, FET, Manav Rachna Inteational University, Faridabad, India [email protected], dipali. [email protected].in. Abstract- Electrocardiogram (ECG) is a primary clinical diagnostic tool for detection of cardiac arrhythmias. As ECG signals are generally acquired over longer time periods at extremely high resolution and thus are highly data intensive. This leads to the requirement of large storage space for database construction and more transmission bandwidth for remote ECG signal analysis, respectively. Successive ECG beats and sample values however, show some redundancy along with the information content. By removing this redundancy, ECG signal compression can be achieved. This paper comprises implementation of hybrid ECG signal compression system based on frequency transformation and parameter extraction techniques. It uses discrete cosine transform (DCT) and Fast Fourier transform (FFT) to compress the ECG signal. This compressed ECG signal is embedded with corresponding heart rate information in order to obtain a high quality reconstructed signal required for accurate cardiac state diagnosis. The proposed algorithm is tested for compression of bradycardia and tachycardia ECG rhythms selected from MIT-BIH arrhythmia database and the performance is evaluated using compression ratio and percent root-mean-square difference (p). The high compression ratio, low reconstruction error and less computational complexity justify the efficiency of hybrid techniques in ECG signal compression and thus, in telecardiology. Index Terms-ECG compression, compression ratio, P, discte cosine transform (DCT), Fast Fourier transform (FFT), heart rate. I. INTRODUCTION Long-term ECG signal recording and analysis is a primary step towards cardiac state recognition. Computerized ECG monitoring systems are generally, installed for continuous acquisition and digital storage of sUbject's cardiac records. This requires large storage space and large transmission bandwidth for remote ECG signal analysis. Aim is towards development of an efficient ECG signal storage and ansmission system. Compression ofECG data reduces the number of bits required to represent the cardiac signal while retaining the important diagnostic information in the reconstructed signal [1]. Successive ECG beats and sample values show some redundancy along with the information content. By removing this redundancy, ECG signal can be compressed. In the recent years, number of ECG signal compression algorithms has been introduced. Earlier direct signal compression methods were used that involve the direct compression of signal samples in the time-domain itself The signal is reconsucted by interpolation of extracted adjacent samples. This category includes AZTEC [2], CORTES [3], 978-1-4799-2995-5/14/$31.00©2014 IEEE 437 FAN/SAPA [4], ing Point [5], SLOPE algorithms [6] and Delta coding [7]ECG compression schemes. A nother category consisting of distinct parameter exaction methods involves extraction of some particular parameter of theECG signal to be compressed. This parameter itself is used to represent signal information and the consequent signal reconsuction is performed using spline nctions. Algorithms such as long term prediction (LTP) [8] and peak picking [9] belongs to this category. The most widely used ECG signal compression algorithms are based on equency transformation techniques. first transforms ECG signals to equency domain to compress the signal followed by computation of inverse transformation in order to reconsuct the ECG signal with acceptable distortion levels. This category includes Karhunen-Loeve transform (KLT) [10], cosine transforms (CT) [11], Fourier transform (FT) [12], Walsh transform (WT) [13] and recent wavelet transform ( cWT/DWT) [1]. [14]-[17]. Most of the above stated methods do not permit efficient reconstruction of the signal and thus fail to preserve significant diagnostic features ofECG signals. Thus, aim is towards the development ofECG signal compression technique that achieves desired reduced information content while retaining the significant diagnostic features ofECG signals In this research, a equency transformation and parameter extraction based hybrid ECG signal compression system is developed. At first, the transformed coefficients are obtained by applying discrete cosine transform (DCT) and Fast Fourier transform (FFT) to the input ECG signal that in t are embedded with the coesponding heart rate information in order to reconsuct a hi-fidelity ECG signal. DCT being a real transform and FFT being a fast algorithm, are computationally efficient transforms. II. MATEALSAND METHODS A nctional block diagram of the hybrid ECG signal compression system based on equency ansfoation and parameter extraction techniques is presented in Fig. 1. The whole methodology for hybrid ECG signal compression system is implemented in three major steps namely pre- processing, DCT/FFT transformation and coding the transformed coefficients with heart rate information.

Transcript of [IEEE 2014 International Conference on Optimization, Reliabilty, and Information Technology (ICROIT)...

Page 1: [IEEE 2014 International Conference on Optimization, Reliabilty, and Information Technology (ICROIT) - Faridabad, Haryana, India (2014.02.6-2014.02.8)] 2014 International Conference

2014 International Conference on Reliability, Optimization and Information Technology -

ICROIT 2014, India, Feb 6-8 2014

Hybrid ECG Signal Compression System: A Step Towards Efficient Tele-Cardiology

Rashima Mahajan, Dipali Bansal Research Scholar, Dept. ofEEE , FE T, Manav Rachna International University, Faridabad, India

Adjunct Faculty, Dept. of ECE , SE T, Apeejay Stya University, Gurgaon, India Professor, Dept. ofEEE , FET, Manav Rachna International University, Faridabad, India

[email protected], dipal i. [email protected].

Abstract- Electrocardiogram (ECG) is a primary clinical diagnostic tool for detection of cardiac arrhythmias. As ECG signals are generally acquired over longer time periods at extremely high resolution and thus are highly data intensive. This leads to the requirement of large storage space for database construction and more transmission bandwidth for remote ECG signal analysis, respectively. Successive ECG beats and sample values however, show some redundancy along with the information content. By removing this redundancy, ECG signal compression can be achieved. This paper compr ises implementation of hybrid ECG signal compression system based on frequency transformation and parameter extraction techniques. It uses discrete cosine transform (DCT) and Fast Fourier transform (FFT) to compress the ECG signal. This compressed ECG signal is embedded with corresponding heart rate information in order to obtain a high quality reconstructed signal required for accurate cardiac state diagnosis. The proposed algorithm is tested for compression of bradycardia and tachycardia ECG rhythms selected from MIT-BIH arrhythmia database and the performance is evaluated using compression ratio and percent root-mean-square difference (pRD). The high compression ratio, low reconstruction error and less computational complexity justify the efficiency of hybrid techniques in ECG signal compression and thus, in telecardiology.

Index Terms-ECG compression, compression ratio, PRD, discrete cosine transform (DCT), Fast Fourier transform (FFT), heart rate.

I. INTRODUCTION Long-term ECG signal recording and analysis is a primary

step towards cardiac state recognition. Computerized ECG monitoring systems are generally, installed for continuous acquisition and digital storage of sUbject's cardiac records. This requires large storage space and large transmission bandwidth for remote ECG signal analysis. Aim is towards development of an efficient ECG signal storage and transmission system. Compression ofECG data reduces the number of bits required to represent the cardiac signal while retaining the important diagnostic information in the reconstructed signal [1]. Successive ECG beats and sample values show some redundancy along with the information content. By removing this redundancy, ECG signal can be compressed.

In the recent years, number of ECG signal compression algorithms has been introduced. E arlier direct signal compression methods were used that involve the direct compression of signal samples in the time-domain itself. The signal is reconstructed by interpolation of extracted adjacent samples. This category includes AZTEC [2], CORTES [3],

978-1-4799-2995-5/14/$31.00©20 14 IEEE 437

FAN/SAPA [4], Turning Point [5], SLOPE algorithms [6] and Delta coding [7] ECG compression schemes. Another category consisting of distinct parameter extraction methods involves extraction of some particular parameter of the ECG signal to be compressed. This parameter itself is used to represent signal information and the consequent signal reconstruction is performed using spline functions. Algorithms such as long term prediction (LTP) [8] and peak picking [9] belongs to this category.

The most widely used ECG signal compression algorithms are based on frequency transformation techniques. It first transforms ECG signals to frequency domain to compress the signal followed by computation of inverse transformation in order to reconstruct the ECG signal with acceptable distortion levels. This category includes Karhunen-Loeve transform (KLT) [10], cosine transforms (CT) [11], Fourier transform (FT) [12], Walsh transform (WT) [13] and recent wavelet transform (cWT/DWT) [1]. [14]-[17]. Most of the above stated methods do not permit efficient reconstruction of the signal and thus fail to preserve significant diagnostic features ofECG signals. Thus, aim is towards the development ofECG signal compression technique that achieves desired reduced information content while retaining the significant diagnostic features ofECG signals

In this research, a frequency transformation and parameter extraction based hybrid ECG signal compression system is developed. At first, the transformed coefficients are obtained by applying discrete cosine transform (DCT) and Fast Fourier transform (FFT) to the input ECG signal that in turn are embedded with the corresponding heart rate information in order to reconstruct a hi-fidelity ECG signal. DCT being a real transform and FFT being a fast algorithm, are computationally efficient transforms.

II. MATERlALSAND METHODS

A functional block diagram of the hybrid ECG signal compression system based on frequency transformation and parameter extraction techniques is presented in Fig. 1. The whole methodology for hybrid ECG signal compression system is implemented in three major steps namely pre­processing, DCT/FFT transformation and coding the transformed coefficients with heart rate information.

Page 2: [IEEE 2014 International Conference on Optimization, Reliabilty, and Information Technology (ICROIT) - Faridabad, Haryana, India (2014.02.6-2014.02.8)] 2014 International Conference

EeG signal Pre-processing DCTIFFT � Compressed Encoded

� database � phase � transformation ECG signal � ECG signal

.. Heart rate ,

information Storagel

a) ECG signal Encoding Transmission

Error � detection

ECG signal �

IDCT/IFFT Decoded -reconstmction transformation

ECG signal I

Cardiac state

diagnosis � Heart rate ., information

b) ECG signal Decoding

Fig. I. Block diagram of the hybrid EeG signal compression system based on frequency transformation and parameter extraction techniques.

A. ECG Signal Pre-processing

The pre-processing step involves the selection and loading ofECG signal records from the MIT-BIH arrhythmia database of Physiobank ATM [18]. The ECG data records representing tachycardia and bradycardia rhythms are selected and exported to MATL AB workspace using MATL AB function: plotATM(,223m.mat', '223m. info'). The detailed process of loading ECG signal records from Physiobank ATM to MATLAB workspace is described in our previous work [19].

B. DCTIFFT Transformation

The pre-processed ECG signal is transformed to frequency domain by applying DCT/ FFT to the signal using MATLAB functions 'dct' and 'fft'. The transformation operation actually aims at de-correlating the original ECG signal information and transforming large information content present in the original signal into a relatively smaller set of transformed coefficients as these techniques possess strong energy compaction property [20]. Thus, leads to compression of the original ECG signal.

C. Coding the transformed coefficients

The encoded ECG data is formed by embedding heart rate information with DCT/ FFT transformed coefficients in order to obtain a high quality reconstructed signal required for accurate cardiac state diagnosis. Fig. 2 details the steps for heart rate estimation and the corresponding waveforms are plotted in Fig. 3 . Heart rate is an indicator of heart beats per minute. The input ECG signal record loaded from the database is filtered using high pass FIR filter using the MATLAB functions : h=fir1(lOOO,1I1000*2,'high') ;

ECG_filter=filter (h,1,ECG) ;

438

The filtered signal is squared in order to obtain high peaks with relatively increased size. Consequently, the R- peaks are detected by comparing the adjacent sample values above threshold. Once the R-peaks are detected, heart rate information is estimated by calculating the number of beats per minute. Ultimately, the encoded ECG data stream is constructed by the fusion of compressed ECG bit stream and associated heart rate information. During decoding process, the signals are reconstructed in order to extract significant ECG signal parameters for diagnostic purposes. The compressed ECG bit stream contains clinically significant information for regenerating each heart beat and thus forms the major part of the encoded data for ECG signal reconstruction using inverse discrete cosine transform (IDCT) and inverse Fast Fourier transform (IFFT). The MALAB functions 'idct' and 'ifft' are used for inverse transforms implementation. The detailed methodology for hybrid ECG signal compression is depicted in the flow chart shown in Fig. 4.

Page 3: [IEEE 2014 International Conference on Optimization, Reliabilty, and Information Technology (ICROIT) - Faridabad, Haryana, India (2014.02.6-2014.02.8)] 2014 International Conference

Load ECG signal to MATLAB from

MIT-BIH arrhythmia database.

Set FIR digital filter parameters.

Set high pass filter frequency.

Plot filtered ECG signal.

Obtain squaring of Filtered ECG signal.

Plot squared ECG signal.

Set threshold.

Heart rate detemlination by detecting peaks.

Fig. 2 Flow chart for determination of heart rate information

: I • • • _ _ _ _ _

: -"" ....

-""""

Samples

Fig 3. ECG waveforms during heart rate estimation.

Evaluate DCT compression by

CR and PRD detenninatioo.

Evaluate FFT compression by

CR and PRD detenmllation.

Fig.4. Detailed flow chart for hybrid ECG signals compression system

439

D. Performance Evaluation

The performance of the developed hybrid ECG signal compression system is evaluated using two distortion measures: compression ratio (CR) and percent root mean square difference (PRD) [21], [22]. The compression ratio is defmed as

CR = Nr Nc

(1 )

where NI denote the number of bits representing the input ECG signal and NC denote the number of bits representing the compressed ECG signal . After decompression, the reconstructed signal is obtained. The distortion between the original and corresponding reconstructed ECG signal is evaluated in terms of percent root mean square difference (PRD) error index. The PRD is calculated using,

PRD= (2)

where x [n] represents the original signal,:i nl represents the reconstructed signal and N, the number of samples. Higher the value of CRt higher is the degree of compression while low PRD indicates the efficient reconstruction of ECG signal, thereby preserving the significant clinical information even after compression.

III.RESULTS AND DISCUSSION

This section describes and analyzes the results of ECG signal compression that verify the effectiveness of the frequency transformation based ECG signal compression system. The algorithm is developed and implemented using MATLAB 7.10 on Core i3 processor. The developed algorithm is tested and evaluated using ECG records extracted from MIT­BIH arrhythmia database of Physiobank ATM. The records selected for test dataset are 223 and 232 corresponding to tachycardia and bradycardia rhythms, respectively. The DCT compression results for the two datasets consisting of first 4000 samples ofECG records 223 and 232 are shown in Figs. 5(a) and 5(b), respectively. Similarly, the FFT compression results for the same datasets are shown in Figs. 6(a) and 6(b), respectively. The signal is reconstructed by applying inverse discrete cosine transform and inverse Fast Fourier transform to the DCT compressed and FFT compressed signal, respectively. The reconstructed signal is not exactly same as the original one and the error is indicated by PRD. FFT compression algorithm provides higher degree of compression but with higher PRD' s as compared to DCT as depicted in Table I. As low PRD indicates the efficient reconstruction of ECG signal thus, DCT provides better signal reconstruction compared to FFT. Further DCT being a real transform, is computationally more efficient than FFT.

It has been noted that the performance of the compression algorithm is also dependent on the ECG record being compressed. The compression ratio achieved is higher for bradycardia ECG record (CRDCT=94.60,CRFFT=98.22) as compared to tachycardia (CRDCT=89 .62,CRFFT=91.32)with both DCT and FFT compression algorithms. On the other hand, the PRD achieved is lower in case of tachycardia (PRDDCT=0.58%, PRDFFT=1.08%)as compared to

Page 4: [IEEE 2014 International Conference on Optimization, Reliabilty, and Information Technology (ICROIT) - Faridabad, Haryana, India (2014.02.6-2014.02.8)] 2014 International Conference

bradycardia (PRDDCT=2.11%, PRDFFT=3.55%). The performance results in terms of CR and PRD for compressing both the datasets using DCT and FFT are plotted in Fig. 7( a) and (b), respectively. The developed hybrid ECG signal compression system shows superior performance as compared to other compression algorithms as stated in the literature.

on parameter extraction and frequency transformation techniques DCT and FFT has been developed and tested using bradycardia and tachycardia ECG rhythms extracted from MIT-BIH arrhythmia database. Heart rate information along with the compressed ECG bit stream constitutes encoded ECG signal. High compression ratio while retaining low percent root mean square errors are achieved indicating a high-quality reconstruction of compressed ECG signals. It enables cardiologist to accurately analyze and detect cardiac condition of the patient by extracting important diagnostic parameters from the reconstructed ECG signal. Thus, frequency transformation techniques DCT and FFT possess the potential to be used for compression of ECG signals. This provides feasibility towards efficient ECG signal storage and transmission system for long-term ECG database construction and tele-cardiology in emergency conditions, respectively. Efforts are also underway to inspect quality of compression and reconstruction by estimating the peak signal-to-noise ratio of ECG signal.

ECG signal record

223 232

Table I. Performance Of The Hybrid Ecg Compression System With Oct And Fft

DCT compression FFT compression

CR PRD(%) CR PRD (%) 89.62 0.58 9l.32 1.08 94.60 2.11 98.22 3.55

IV. CONCLUSION

An efficient hybrid ECG signal compression system based

Original ECG signal

.� o 500 HID 1500 21m � 3()JJ 350J 4()JJ

OCT encoded E CG signal

500 l00J 1500 2500 OCT encoded E CG signal

500 l00J 1500 21m 2500 3()JJ Reconstrucled ECG signal using IDCT

.� o 500 1 ()JJ 1500 21m � 3()JJ 3500 4()JJ

Error signal

.::��-��\��·�-�'�7�-'-�·===:1 o 500 l00J 1500 21m 2500 3()JJ 3500 400J

HID

50) 1COl

1500

Original ECG signal a)

1500 OCT Mcoded ECG signal

1500 Reconslructed ECG signal using IDCT

21m ElTl)r signal

b)

25CJO

21m 25CJO

3500

Samples

Fig. 5 Compression analysis of ECG signals records using OCT. Each figure shows the original, OCT compressed, reconstructed and error signals.

400J

a) The first 4000 samples ofMIT-BIH record 223, CR=89.62, PRD=0.58%. b) The first 4000 samples of MlT-BIH record 232, CR=94.60, PRD=2.11 %.

440

Page 5: [IEEE 2014 International Conference on Optimization, Reliabilty, and Information Technology (ICROIT) - Faridabad, Haryana, India (2014.02.6-2014.02.8)] 2014 International Conference

Origin.1 ECG sign. I

.� o 500 1 COl 1500 2COJ 2500 3(OJ 3500 4COJ FFT .neoded ECG signal

���������������� ·�O�--------�500�-----------ICOI�----------�15OO�----------2COJ��--------� 2500� �--------�3(OJ� ----------�3500�----------�4COJ·

FFT .neoded ECG signol

500 HXXl 1500 2COJ 2500 3C(() 3500 Reconsl,uol.d ECG signol using Invers. FFT

.� � o 500 1C01 1500 2COJ 2500 3(OJ 3500 4C01

Error signal

a)

O,igin.1 ECG sign.1

.�fl-------------��---------: �-��--r-------f o SIll lIDJ ISIlI 2IlD 2500 3(OJ 3500

FFT encoded ECG sign.1

.��-----7.7-----��----��-----=�-----=�----��----��----�-' o SIll 11))] 1 SIll 2IlD 2500 DXl 3500 400J

SIll 11))]

FFT encoded ECG sign.1

IS1l1 2IlD 2500 ReconslruclBd ECG sign. I using l"",r'B FFT

oj 3500 400J

.H � � �--�-: -3 o SIll 11))] ISIlI 2IlD 2500 DXl 3500 400J

Error sign.1

7111 IlOJ lID)

Samples b)

Fig. 6 Compression analysis of ECG signals records using FFT. Each figure shows the original, FFT compressed, reconstructed and error signals. a) The first 4000 samples of MlT BlH record 233,CR=91 .32, PRD=I.08%.b) The first 4000 samples of MIT BlH record .232, CR=98.22, PRD=3.55%.

4 100

3.5 98 � 96 .. c:: c 94 0

.;;; '" 92 .. ...

3 ,-.. � 2.5 � 0 2 • DCT c:: �

c. E 90 • FFT 1.5 0 U 88 1

86 0.5

84 0

Record 223 Record 232 Record 223 Record 232

ECG signal records from MIT-BIH arrhythmia database. ECG signal records from MIT-BIH arrhythmia database.

a) b) Fig.7. Performance analysis of ECG signals compression in terms of a) CR and b) PRD, using both DCT and FFT.

441

Page 6: [IEEE 2014 International Conference on Optimization, Reliabilty, and Information Technology (ICROIT) - Faridabad, Haryana, India (2014.02.6-2014.02.8)] 2014 International Conference

REFERENCES

[I] B. A. Rajoub, "An efficient coding algorithm for the compression of ECG signals using the wavelet transform", IEEE Trans. Biomed. Eng., Vol. 49, No. 4, pp. 355-362, April 2002.

[2] l R. Cox, F. M. Nolle, H. A. Fozzard, and G. C. Oliver, "AZTEC, a preprocessing program for real-time ECG rhythm analysis", IEEE Trans. Biomed. Eng., vol. BME-15, pp. 128-129,ApriI 1968.

[3] l P. Abenstein and W. l Tompkins, "New data-reduction algorithm for real-time ECG analysis", IEEE Trans. Biomed. Eng., vol. BME-29, pp. 43-48, Jan. 1982.

[4] J . Sklansky and V. Gonalez, "Fast polygonal approximation of digitized curves",Pattern Recog., vol. 12, pp. 327-331, 1980.

[5] W. C. Mueller, "Arrhythmia detection program for an ambulatory ECG monitor", Biomed. Sci. Instrument., vol. 14, pp. 81-85, 1978.

[6] S.c. Tai, "SLOPE- A real time ECG data compressor", Int. l Bio-Med.and Computers., vol. 29,pp. 175-179, 1991.

[7] D. Stewart, G. E. Dower, and O. Suranyi, "An ECG compression code", l Electrocardiol., vol. 6, no. 2, pp. 175-176,1973.

[8] G. Nave and A. Cohen, "ECG compression using long-term prediction", IEEE Trans. Biomed. Eng., vol. 40, pp. 877-885, Sept. 1993.

[9] A. Cohen, P. M. Poluta, and R. Scott-Millar, "Compression of ECG signals using vector quantization," in Proc. IEEE-90 S. A. Symp. Commun. Signal Processing COMSIG-90, Johannesburg, South Africa, pp. 45-54, 1990.

[10] M. E. Womble, l S. Halliday, S. K. Mitter, M. C. Lancaster, andl H. Triebwasser, "Data compression for storing and transmitting ECGs/vCGs", Proc. IEEE, vol. 65, pp. 702-706, May 1977.

[II] N. Ahmed and K. R. Rao, Orthogonal Transforms for Digital Signal Processing. New York: Springer, 1975.

[12] B. R. S. Reddy and I. S. N. Murthy, "ECG data compression using Fourier descriptors", IEEE Trans. Biomed. Eng., vol. BME- 33, pp. 428-434, Apr. 1986.

[13 ]w. S. Kuklinski, "Fast Walsh transform data-compression algorithm; ECG application", Med. BioI. Eng. Comput., vol. 21, pp.465-472,July 1983.

[14] Z. Lu, D. Y. Kim, and W. A. Pearlman, "Wavelet compression of ECG signals by the set partitioning in hierarchical trees (SPIHT) algorithm", IEEE Trans. Biomed. Eng., vol. 47, pp. 849-856, July 2000.

[15] S. C. Tai, C. C. Sun, and W. C. Yan, "A 2-D ECG Compression Method Based on Wavelet Transform and Modified SPIHT", IEEE Transactions on Biomedical Engineering, vol. 52 (6), 2005.

[16] A. Alesanco and l Garcia, "A simple method for guaranteeing ECG quality in Real-time Wavelet lossy coding", EURASIP Journal on Advances in Signal Processing, vo1.2007, pp. 1-9, 2007.

[17] S.M. Ahmed, Q. AI-Zoubi and M. Abo-Zahhad, "A hybrid ECG compression algorithm based on singular value decomposition and discrete wavelet transform," l Med. Eng. Technology, vol. 31, pp. 54-61,2007.

[18] MIT-BIH Database distribution, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, I 9 9 8 http://www.physionet.org/physiobankldatabase/mitdb/.

[19] R. Mahajan and D. Bansal, "Compact Feature Vector based ECG Beat Classification using Neural Network Classifier", in proceedings of International Conference CERA-2013, pp. 290-295, lIT Roorkee, October 20 13.

[20] W.B. Pennebaker and lL. Mitchell, "JPEG Still Image Data Compression Standard", New York, NY: Van Nostrand Reinhold, 1993. Chapter 4.

[21] M. B. Velasco, F. C. Roldan, l I. Llorente, and K. E. Barner,

442